少标签下扩散模型与半监督学习器的互惠互利
Diffusion Models and Semi-Supervised Learners Benefit Mutually with Few Labels
AI总结:
提出 DPT 三阶段训练策略,通过伪标签与扩散模型生成伪图像并反哺分类器,在少标签 ImageNet 生成和分类上达到 SOTA。
摘要:
为进一步推进半监督生成与分类任务,我们提出一种简单而有效的训练策略,称为 dual pseudo training(DPT,双重伪训练),它构建于强大的半监督学习器和扩散模型之上。DPT 分为三个阶段:在部分标注数据上训练分类器以预测伪标签;使用这些伪标签训练条件生成模型,以生成伪图像;以及用真实图像和伪图像的混合数据重新训练分类器。实验上,DPT 在多种设置下持续取得半监督生成与分类的 SOTA 性能。具体而言,在每类一个或两个标签时,DPT 在 ImageNet 256×256 上取得 3.08 或 2.52 的 Fréchet Inception Distance(FID)分数。此外,DPT 在 ImageNet 分类任务上显著优于具有竞争力的半监督基线,在每类一个、两个或五个标签时分别取得 59.0(+2.8)、69.5(+3.0)和 74.4(+2.0)的 top-1 准确率。值得注意的是,我们的结果表明,扩散模型仅用极少标签(例如 <0.1%)即可生成逼真图像,并且生成式增强对于半监督分类仍然可行。代码位于 https://github.com/ML-GSAI/DPT。
英文摘要:
In an effort to further advance semi-supervised generative and classification tasks, we propose a simple yet effective training strategy called dual pseudo training (DPT), built upon strong semi-supervised learners and diffusion models. DPT operates in three stages: training a classifier on partially labeled data to predict pseudo-labels; training a conditional generative model using these pseudo-labels to generate pseudo images; and retraining the classifier with a mix of real and pseudo images. Empirically, DPT consistently achieves SOTA performance of semi-supervised generation and classification across various settings. In particular, with one or two labels per class, DPT achieves a Fréchet Inception Distance (FID) score of 3.08 or 2.52 on ImageNet 256x256. Besides, DPT outperforms competitive semi-supervised baselines substantially on ImageNet classification tasks, achieving top-1 accuracies of 59.0 (+2.8), 69.5 (+3.0), and 74.4 (+2.0) with one, two, or five labels per class, respectively. Notably, our results demonstrate that diffusion can generate realistic images with only a few labels (e.g., <0.1%) and generative augmentation remains viable for semi-supervised classification. Our code is available at https://github.com/ML-GSAI/DPT.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。